Written by Clay Smith
Spoon Feed
The win ratio can be a useful study design that allows hierarchical ranking, by clinical relevance, of composite outcomes into a win, tie, or loss, but watch out for abuses of it. They can be “seductive.”
It’s a win…ratio
We covered an article that not intubating intoxicated patients with GCS≤8 was better than usual care. This study used a somewhat novel (at least to me) primary outcome called the win ratio. A composite outcome equally weights all components, i.e. death = ICU LOS = hospital LOS. Clearly, death is worse than ICU LOS. Whereas a win ratio allows hierarchical ranking, by clinical relevance, of composite outcomes into a win, tie, or loss. Each patient is compared to every other patient in the other group to determine win, tie, or loss. Two recent articles came out in different journals; one with pro/con and the other focusing only on the drawbacks.
Pro
Here are some upsides to using a win ratio.
- In a typical analysis using logistic regression or Cox proportional hazard analysis, the hierarchical nature of outcomes by clinical importance could be lost. See image. For example, patient A is dead but would appear to have a “superior” outcome to patient B because death occurred later. That seems kind of…wrong, doesn’t it?
- In critical care trials, a win ratio can still assess “hard outcomes,” such as mortality, but can also allow for consideration of other clinically important, patient-centered outcomes as well. This may provide a more well rounded understanding of the overall effect of treatment.
- Since each patient in the intervention is compared to each control (i.e. intervention N x control N), the overall sample size for a trial can often be reduced. This increased power can also detect smaller, clinically important differences in trials with a moderate sample size.
Con
These authors emphasize the downsides of a win ratio…calling it “seductive,” which I find hilariously nerdy.
- When using a win ratio, it can be tough to judge which variables are worse than others; which is worse, MI or stroke, dialysis or increased ICU LOS?
- It could fail to account for long-term events. Sometimes major, important outcomes may not occur until later, so you could log an early “win” but fail to detect a later important “loss.”
- As mentioned, a win ratio allows greater power with a smaller sample size, but that may simply show statistically but clinically insignificant differences in an outcome.
- Ties are ignored, and they make an excellent point about this: “Ties represent the absence of a ‘win,’ and ignoring them exaggerates the treatment effect.” True.
How will this change my practice?
We need to understand the “damn lies” in all statistical methods, including the win ratio. Be wary if the percentage of “ties” is high. Make sure you agree on how the author ranked variables by clinical importance. And trash a paper if the win ratio is applied post hoc.
Sources
Use of the Win Ratio Analysis in Critical Care Trials. Am J Respir Crit Care Med. 2024 Apr 1;209(7):798-804. doi: 10.1164/rccm.202309-1644CP. PMID: 38285595
Win Ratio: A Seductive But Potentially Misleading Method for Evaluating Evidence from Clinical Trials. Circulation. 2024 May 14;149(20):1546-1548. doi: 10.1161/CIRCULATIONAHA.123.067786. Epub 2024 May 13. PMID: 38739696
